GEOPHYSICS, VOL. 34, NO. 1 (FEBRUARY 1969), P. 39-53, 9 FIGS.
, 1 TABLE
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
THE
C. N. G. DAMPNEY*
The inherent ambiguity of potential field inter-
pretation can be put to advantage. Bouguer
anomaly measurements on an irregular grid and
at a variety of elevations can be synthesized by
an equivalent source of discrete point masses on a
plane of arbitrary depth below the surface. By
keeping the depth of the plane within certain
limits relative to the station spacing, we can
ensure that the synthesized field closely approxi-
mates the true gravity field in the region close to
fundamental property of a potential field is A
the inherent ambiguity in determining its source
from values of the field outside the source region.
For instance, given the vertical anomalous gravi-
tational field intensity, g, (x, y, z) over a hori-
zontal plane, it is impossible to find the anomalous
mass distribution uniquely.
Roy (1962) has discussed in detail the ambi-
guity of the relationship between g, (x, y, z) and
c (c~,/3, h) in the equation
where u ((Y, p, h) is the surface contrast density
distribution over the plane z= h, K is the gravita-
tional constant and the positive direction is
down. We note that the depth h of the apparent
source may take any value.
To perform the inversion of equation (1.1) it is
sufficient to have a complete knowledge of g, on a
plane z=zr zi <h. Therefore a unique corre-
spondence exists between the function g, (x, y,
z=zi) and (T (x, y, h) ; and furthermore as a result
t Presented at the 37th Annual International SEG Meeting, Oklahoma City, Oklahoma, November 1, 1967.
Manuscript receivedby the Editor December4, 1967;revised manuscriptreceivedOctober 30, 1968.
* Dept of Geology,University of Tasmania, Australia; Contribution No. 187. Now with GeophysicsLaboratories,
University of Toronto, Canada.
EQUIVALENT SOURCE TECHNlQlJEt
and above the terrain.
Once the equivalent source is obtained, the
projection of the Bouguer anomaly onto a reg-
ularly gridded horizontal plane is easily done.
In addition, the equivalent source can then be
economically used to carry out vertical continua-
tion. The technique is illustrated by a hypothet-
ical example and a case history of a local gravity
survey in precipitous topography.
INTRODUCTION of equation (1.1) any value of g, (x, y, 2) for z <h
is similarly defined uniquely.
Thus we fiild a surface density contrast distri-
bution on an arbitrary plane which synthesizes a
known gravity field. It is then possible to cal-
culate from gravity measurements at one height
the gravity field at any other point in space.
The projection and interpolation of the field from
known points of measurement may thus be
carried out.
gz(x,y, 2) = K
J --oo J --oo I<%- o)* + (y - P)” + (2 - h)Zj3’?
By using the equivalent source u (a, p, h) as an
integral part of gravity field computations, two
advantages result: (1) All available information
contained in the measurements of g, (x, y, z) is
used, and (2) we ensure the correct analytical
gravity potential field behavior of the projected
field values intrinsically as they are calculated
from a gravitating equivalent source.
39
 40 Dampney

REVIEW uous surface density contrast distributiona(c+,h)

There are, in the literature, many data-pro- on any plane zl=h. The equivalent source
cessing techniques (e.g. Henderson, 1960; Bhat- technique is based on approximating this con-
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tacharyya, 1965, 1966; Keidell, 1966; Sax 1966) tinuous distribution by a series of discrete masses.
which require data to be put on a gridded hori- If we have N data points, we can calculate the
zontal plane. In practice, however, points of values of N point masses at a suitable depth which
measurement may be scattered horizontally and will then constitute the equivalent source from the
vertically over normal rough topography. equations (using the principle of superposition).
i\s Naudy and Neumann (1965) emphasize, the
gl = aim + 61121132 + . .
Bouguer anomaly must be regarded as being the
vertical gravity field at points on the terrain due
+ alkmk + . + (IlNmN
to the anomalous massesin the ground. The points
are not situated on the geoid or some other g2 = azlml + a2?m2 + . . .
reference surface. The observed values must not
only be interpolated horizontally from their orig- + a?kmk + ’ ’ ’ + a?NmN

inal coordinates of measurement, but be ver- . . . . . . . . . . . . . .

. (3.1)
tically projected onto the plane of the grid as
well. gi = Uii?YZi+ Uigm2 + . . .
Projection of Bouguer anomalies onto a hori-
+ aikmk + . . . + Ui.vWZ.v
zontal plane has received very little attention in
the literature. Strakhov and Devitsyn (1965) and . . . . . . . . . . . . . . .
Grant and West (1965, p. 269) have proposed the
only two solutions known to the author. In this
paper we set up an equivalent source to represent
and thereby interpolate and project the observed
Bouguer anomaly values.

uik
= {(xi- cYk)*+ (ri - @k)*+ (zi - /4*)3/z

The idea of an equivalent layer has been ex-
ploited by Danes (1961) in gravity interpretation
incorporating borehole data and recently by Bott
(1967) to interpret magnetic anomaly profiles.
This demonstrates that representing potential
fields by an equivalent source may have many
applications. In fact, as Zidarov (1965) showed,
the idea can be applied to electrical fields and to
the general Dirichlet-Neumann inverse problem
as well as to gravity and magnetics.
Zidarov’s (1960, 1965) papers give the general
outline of representing potential fields by equiva-
lent sources.
THE EQUIVALENT SOURCE TECHNIQUE
From equation (1.1) we see that the anomalous
gravitational field intensity g, (x, y, z) can be
represented by or be synthesized from a contin-
and z=h is the horizontal plane containing the
point masses mk at ((Yk,@k, h). The position of g,
is (Xi, Yz, &I.
This can be written in matrix form
g = Am (3.3)
which represents N simultaneous equations in N
unknowns and is thus solvable.
Suppose the N data points gi have an average
station spacing of Ax. The approximation of the
discrete masses mk to a continuous distribution
will be a valid representation of g if the mk are
sufficiently far below the surface relative to Ax.
Make the average separation of the N masses
AZ so as to correspond to the average station
spacing of gi. Consider the anomaly gz (x, y, s)
at (x, y, z) resulting from the discrete masses
mk at (ffk, Pk, h)

N
mk(h- 4
k=l ((x - ak)2 + (y - ok)’ + (2 - h)2f3’2 ’
 Equivalent Source Technique 41

Taking the Fourier transform of g, we obtain:

G,(u, z’, z) = 27rK esp - e 4-x)’ + (zlAx); \I X 2 mkek, (3.5)

Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

I k=l

where tion of a gravity field, due to a line source, below

the source depth induced violent oscillations in
ek = exp (-&U - &v)
the field. As the gravitational intensity and sur-
and face contrast density are linked by the equation

u = 21rjz; v = 21rjU,

where the fz and .f, are spatial frequencies mea-

sured in cpm in the mks system. the equivalent source would also experience this
effect if placed below the level of a point or line
1 CwkckI 2 c mk as / ck 1 2 1. (3.6)
source. In the synthetic example following, the
Hence G, (p, Y, z) is asymptotically dominated by depth of the equivalent source was taken at two-
the term thirds the depth of the anomalous gravitating
sphere.
In the case of a local survey, its area1 extent
may also limit the depth of the plane Zi. If (h-z<)
is large relative to the dimensions of the survey,
.d[(uAx):! + (vAx)~] . (3.7) the coefficients sib tend to approach a common
1 value a, where

(h - zi)
a=
I: ((Xi - (.yk)’ + (y; - /3k)* + (zi - A)“] 312

= im l/(z - h)*. (3.9)

h+co

hccording to sampling theory (Blackman and Thus the matrix .4 becomes ill-conditioned and
Tukey, 1959) the maximum frequency at which its solution unreliable if the equivalent source is
Gz(u, 7’, z) can be specified from the set of values too far below the surface; that is if
g, (xi, yi, z;) is the aliasing or folding frequency
(h - Zi)! 1(Xj - (ok)’ + (yi - fik)‘) “2
given by urnax= r/Ax. In fact, in a well-designed
survey, the amplitude of G,(u, D, z) computed is too large in equation (3.9).
from the g, will approach zero at the aliasing fre- From equation (3.5), a lower limit of (h-z,) =
quency. 2Ax gives
Hence, in spectral terms, the discrete equiv-
alent source representation of gz must also (unl,,, X’lKl,, z) = 2 X lo-* G(0, 0, z) (3.10)
satisfy this requirement. From equation (3.7) sufficient to make G negligible beyond the aliasing
Gz(u, V, z) is seen to be negligible at frequency.
rt,,,, = x/Ax; Tln.1,= T/AK In a test on part of the data in the case history
(3.8)
following, it was found that a value of (h-zi)
providing (h-z)/Ax is sufficiently large. This ?SAx (Table 1) did not make the matrix A ill-
condition places an upper limit on the plane zi of conditioned, demonstrating that the limits on
the equivalent source. (h-ZJ are sufficiently broad to cover the case of
;\n effect which places a lower limit on the rough topography. Over the entire survey (h-
plan: zi can be deduced from Bullard and Cooper zi)=2.5Ax.
(1948). They noted that the downward continua- In summary the values (/z-z,) should satisfy
 42 Dampney

2.5Ax < (h - zi) < 6Ax, (3.11) forces the solution through regions in hyperspace
where the hyperellipsoid of R becomes very
where the upper bound is based empirically on the elongated. This makes the method over-relax
case history.
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

raising the possibility of a divergent iteration, in

Equation (3.1) was solved by using Zidarov’s
application of the “method of steepest descent”
which is based on the geometric notion that the
equation can be solved by minimizing R with
respect to mk.
R = (g - Am)r(g - Am),
where the superscript T denotes the transpose
operator.
This is done by choosing a unit vector v, along
the line of maximum change R and along which
the initial vector m(l) is moved a maximum dis-
tance x(l) torn@). “Distance” is meant in the sense
of Ralston (1965, page 44) for a hyperspace of
dimension N, where N is the number of variables
m,.
In general
which case the iteration is repeated with succes-
sively halved values of a until it does converge.
While theoretically it is possible to reduce R to
zero, it is more expedient to reduce R to a value at
which it is mainly composed of the random errors
(3.12) in g,. As R follows the path of steepest descent in
hyperspace, the random errors e, of gi do not
influence the gradient vk =dR/&nk or A. In ac-
cordance with normal practice (e.g. Kempthorne,
1962, p. 129) assume e, normal with mean zero
and variance (T.
=ZgeiE+Z$,f:E (3.17)
k k

m(j+l) = m(j) + X(j)v(j)_ where

(3.13)
4‘
Naturally, the reduction of R to zero by m(j) f; = g: - c
would give the solution of equation (3.1). k=l

The geometry underlying this method is dis- mk(h - Zi)

cussed in the appendix together with the deriva-
tion of Zidarov’s equation (2). Dropping the
’ {(xi - ak)* + (yi - @k)’ $ (zi - h)‘) 3’2
superscript (j) we have and g”iis the value of gi stripped of its random
error ei. That is gi= g:+ ei, and therefore /i=
,fi+Ei. As (afi/&&) <0 for all i and k, then

(3.14)

where as N--+a from the property that ei has a zero

mean.
Thus, in the solution of g=Am, the error cri-
terion R must be reduced until it is less than E,
and the total estimated summed, squared noise in the

-
N
mk(h- Zi)
fi = gi - t (3.15)
k-1 ((Xi - Lyk)* f (J’i - Pk)2 + (Zi - h)2j3’*

where (x,, yi, zi) are the points of measurement
and (&, Pk, h) iS the pOSitiOn Of mk.
For practical applications the introduction of a
factor a> 1 into equation (3.13),
m(sti) = m(j) + &,(j) v(j)
measurements of g. This advantage of the tech-
nique in reducing the presence of random errors in
the solution of the mk is passed onto the Bouguer
anomaly values later computed from the equiv-
alent source.
(3.16) E was worked out from
 Equivalent Source Technique 43

A’ Table 1. The numerical convergence of R

R=CfI using the steepest descent procedure
i=l
Valuezf
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Itera-
= g (Ei+fY R=z
tion
cycle
Conver-
gence?
i-1
(8
(gi- gP)
asft+O while R-+0 despite the influence of the
a) Derby-Winnaleah Area (h-z;) =2.5&
random numbers pi, therefore R+xLN, (cJ2 as 3887.459 yes
f&O. 1562.7364 ; yes
Thus there is no point in reducing R below the yes
1063.3290
739.07411 : yes
value Nu2 where c is the variance of errors. The Note slow 556.31239 5 yes
process of iteration should continue until convergence 556.31232 6 yes

Region of 1 2896.7126 7 no
R < E = _17ag. (3.19) nonconvergence\ 556.31234 8 no

Nate fast yes

I3 deveioping an algorithm to solve the matrix convergence yes
[equation (3.1)] by the method of steepest de- yes
scent, the large number of coefficientsaikgenerated yes
44.441891 13 yes
cause computer storage problems. This may be ~ ~~
overcome by either storing the values tempo- b) Test Area (h-z;)=S.OAx
1197.9390 1 yes
rarily outside core-storage if sufficient locations are
yes
not available or by recomputing the coefficients 214.14105
384.77229 : yes
when required. The latter method was chosen in 118.76362 4 yes
Note slow ( 84.163410 - yes
this case. The algorithm requires N2/5000 set per convergence \ 84.163409 :, yes
iteration on an I.B.M. 360/65 where N is the
number of data points. Thus 1000 points would Region of 1 467.92798 7 no
nonconvergence 84.163406 8 yes
require three minutes per iteration. 84.163423 9 no
Table 1 gives the values of R for successive
iterations in analyzing the data discussed in the yes
convergence
Note fast 19.190139
30.586830 :: yes
Case History section below.
The algorithm may be speeded up by improving
the initial approximations to mk. The survey is
broken up into a number of interconnecting Figure 1C compares the theoretically correct
blocks of data sufficiently small so that the ma- field and the field projected from the terrain sur-
trix which solves for their equivalent source fits face values to a common datum plane by use of
within computer core storage. This vastly in- the equivalent source technique.
creases algorithm efficiency as the operation of The difference between the two fields, which
recalculating coefficients is not required. For ex- is the error of the projected field, is plotted in
ample, the equivalent source of 100 points can be Figure 1C. The maximum absolute error, which
calculated in less than a second per iteration on the is seen to occur over the sphere, represents a rela-
360/65. The values of the mk found for the indi- tive error of only three percent.
vidual blocks can then be put together as the
initial values of the equivalent source for the
entire survey.
SYNTHETIC TEST

The technique was tested on synthetic data.

The gravity field of a sphere at a depth of 100
units was computed at points 25 units apart over
an idealized terrain representing a valley and
plateau separated by a 25 unit high cliff (Fig-
FIG. 1A. The synthetic test of an anomalous
ure 1A). This field is contoured in Figure 1B. gravitating body and terrain.
 44 Dam pney

50 100 150 200 250 300 phism. It was extensively intruded and folded by
I I I I I I granite to form hills which rose out of the pre-
5OC
Tertiary landscape. A long period of erosion which
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

followed the granite intrusions, cut deeply into the

topography. Veins in the granite were leached of
tin which was subsequently deposited as low
density (2.0 gm/cm2) alluvium along depressions
in the plane levelled out of the Mathinna Sand-
stone.
\ The topography at the beginning of the Ter-
tiary had thus been formed by the eroding in-
fluence of the Ringarooma River. The river’s
path was controlled by the resistant hills of
granite and sandstone, and by a relative depres-
250 sion in the land surface to the nearby sea which
lowered the river’s slope, changing it to a series of
small lakes and estuaries.
2oc
Then, during the Tertiary, this quiescent scene
was disrupted by the outpourings of basaltic
150 lava flows. The Ringarooma River channel was

IO0

5C

FIG. 1B. The Bouguer anomaly measured over the

terrain of Figure 1A. Contour valuesin mgal.

CASE HISTORY-THE DERBY W’INNALEAH

GRAVITY SURVEY

The equivalent source technique was tested on

data collected by Howland-Rose (1966) on behalf
of the Australian Commonwealth Bureau of
Mineral Resources at Derby, Tasmania, Aus-
tralia. The location of the survey is shown in
Figure 2.
The survey of about 4 sq mi was chosen for this
study as its area is extremely precipitous due to
the influence of the Ringarooma River on the
topography.
The geology of the Ringarooma Valley, which
includes the survey area, has been described by
Nye (1925) (Figure 3).
The oldest rock in the area is the Silurian
Mathinna Group composed of slates and sand-
stones. During the course of geological time this
group has been extensively folded and faulted. FIG. 1C. The absoluteerrorsof the Bougueranomaly
During the Devonian, the Mathinna Group values projected from Figure 1B using the equivalent
sourcetechnique.Dashed contoursrepresentnegative
suffered its most extensive period of diastro- errors.Contour interval= 0.01 mgal.
 Equivalent
BASS STRAIT
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
REFERENCE TO AUSTRALIA t250@0
STANBARD MAP SERIES
FIG. 2. Derby-Winnaleah area, Tasmania, locality map.
buried by lava and its course changed at Derby
as it was forced along the granite basaltic lava
contact in the survey area. The ancient river that
had previously flowed west of the Mt. Cameron
Range (Figure 2) was diverted to the east.
The modern Ringarooma River has eroded its
way down the gorge now containing the Briseis
Mine (Figure 3) in the survey area. A rough
north-south cross section is shown in Figure 4 of a
simplified interpretation of the area’s geology.
The gravity observations were reduced in a
Source Technique 45
SCALE IN M\LES
AREA COVERED 6Y
1964 SURVEY
manner following Hammer (1963) so that the
final topographically corrected Bouguer anoma-
lies took into account the simple model of the
geology shown (see cross section) down to 150
m above mean sea level (“base” level in Figure
4). Bott’s (1961) method for calculating the topo-
graphic effect was employed. The topographic
correction was defined as in Grant and West
(1965, p. 239) and hence, following Bott, the
various densities of the rocks were used to com-
pute the topographic correction. The Bouguer
 46
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

- Elevationcontous AMSL Eevonian Granite 2.6g/cm3 0

80 II II
__-- l&W&tnaC&oup 26g/cm3 m
(inferredfrom air photos)
- Ringorooma River Tertiary Gravek ZOg/cm’O

FIG. 3. Geology and topography of Derby. Elevation values are in units of 100 ft.
Northings and Eastings are in units of 1000 yards.

HILL

PLATEAU
------- 250 METERS
“Y”“YYY”YVY”V
“,“““Y”V” “Y” YYYY ““YV”
Ovvvv BASALTvvv vvv vvv vv
vvYvvYvvv”vvvv V”VVV”V
- ---- yv v v v V_V v-v v-v _v_ S&EgVEL v v 150 METERS
ALLUVIUM
- -~- 3 MILES .-. L

FIG. 4. Diagrammatic cross section through plateau and valley, north to south. All heights
are with reference to mean sea level.
 Equivalent SOI~rce Technique 47

anomalies are shown in Figure 5. As absolute the precipitous topography and the associated
Bouguer anomaly values are not required in a difficulty of making exact topographic correc-
local survey, the zero contour has been set so that tions. The plane of the equivalent source was
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

it passes through the center of the map.
The equivalent source of this Bouguer anomaly
map was found. The noise level parameter E was
worked out from equation (3.19) to be 50 (mgal)*
for the 860 stations involved. The random error of
each Bouguer anomaly value was taken to be the
order of 0.25 mgal. This is reasonable in view of
taken at a height equal to mean sea level which
satisfies the previously discussed limits.
The Bouguer anomalies at regular grid points
on a horizontal plane about the same height as the
basalt plateau (250 m = 800 ft) were then com-
puted from the equivalent source and are plotted
in Figure 6. The grid itself is not shown, but it has

l:IG. 5. TopographicallycorrectedBougueranomaliesfor the Derby-Winnaleah area.

Contour interval = 1.0 mgal.
 48 Dampney
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

FIG. 6. The topographicallycorrected Bouguer anomalies at 250 m above mean sea level.
Contour interval = 0.5 mgal.

a mesh interval of 100 m and extends over the
area contoured. Comparison of Figures 5 and 6
shows that the technique’s projection of the orig-
inal Bouguer anomaly values has eliminated
minor random fluctuations present in the con-
tours of Figure 5. Projection of the values in the
valley region (delimited to the North by the -3.0
contour in Figure 5) to the same reference plane
as the plateau values, allows a direct comparison
of the Bouguer anomalies across the 300 ft eleva-
tion difference between the two regions. Evenly
spaced grid points also allow more objective con-
touring than do long survey lines.
The Bouguer anomalies are now in a satisfac-
tory form for processing by techniques requiring
gridded data.
 Equivalent Source Technique 49

However, the simplicity of directly calculating
the field from the equivalent source suggests us-
ing the technique for vertical continuation. Figure
7 shows the dominant regional trend of the
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
major negative trend across the map apparently
reveals the low density alluvium marking the
buried channel of the ancient Ringarooma River.

The positive anomaly in the southeast corner is

Bouguer anomalies, computed in this way, on a
plane 500 m above sea level.
The final residual map, Figure 8, was found by
subtracting the “regional” values (defined by
Figure 7) from the values at 250 m (Figure 6). The
attributed to the water filled Briseis mine work-
ings taken as being empty in the topographic
corrections. The two positive anomalies in the
northeast and northwest corners, corresponding
to Mathinna sandstone outcrops, mark the limits

\
\
\ \/

FIG. 7. The topographicallycorrectedBougueranomaliesat 500 m above mean sea level.

Contour interval=O.S mgal.
 50 Dampney
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

FIG. 8. The topographicallycorrectedresidualBougueranomaliesat 250 m above mean sea level.

Contour interval=0.2 mgal.

of the alluvial plane imposed by these resistant
pre-Tertiary hills of sandstone and granite.
CONCLUSION
The equivalent source technique as demon-
strated in Synthetic Test section and the Case
History offers a convenient and accurate way to
interpolate gravity data onto a grid. It can be
used to make the final correction to the Bouguer
.anomaly by projecting measurements onto a hori-
zontal reference plane. The technique has the im-
portant application of objective preparation of
Bouguer anomalies for processing by methods
mentioned in the review. It should also be possible
easily to extend this technique to other potential
fields, particularly magnetic fields.
However, limitations have to be realized. Pro-
jecting gravity data onto a horizontal plane in-
volves vertical continuation and so a large hori-
zontal coverage of field values may be required.
 Equivalent
This can be seen from the lateral extent of the
theoretically perfect coefficient set for vertically
continuing a potential field at a height equal to
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
the station spacing. (Dampney, 1966a).
This technique offers a low-error-level, objec-
tive approach to three-dimensional interpolation
and the related problem of contouring data. As
an added bonus it is very economical to use the
equivalent source (once calculated) for reason-
ably accurate vertical continuation.
ACKNOWLEDGEMENTS
This research formed part of an M.Sc thesis
program carried out in the Department of Geol-
ogy, University of Tasmania (Dampney, 1966b).
I gratefully acknowledge the help and inspiring
supervision of Dr. R. Green.
Sincere thanks are extended to Dr. G. F. West,
Geophysics Laboratories, University of Toronto,
for his useful comments made during the prepara-
tion of this manuscript.
The author is also indebted to Mr. J. Boothroyd
for programming advice and to the Director,
Australian Commonwealth Bureau of Mineral
Resources for permission to publish the Derby
Winnaleah Case History.
REFERENCES
Bhattacharyya, B. K., 1965, Two-dimensional har-
monic analysis as a tool to magnetic interpretation:
Geophysics, v. 30, p. 829-857.
- 1966, Continuous spectrum of the total magnet-
ic field anomaly due to a rectangular prismatic body:
Geophysics, v. 31, p. 97-121.
Blackman, R. B., and Tukey, J. W., 19.59,The mea-
surementof power spectra: New York, Dover Pub-
lications,Inc.
Bott, M. H. P., 1961,The useof electronicdigital com-
puters for the evaluation of gravimetric terrain cor-
rections:Geophys.Prosp., v. 7, p. 45-54.
___ 1967!Solutionof the linear inverse problem in
magneticinterpretation with applicationsto oceanic
magneticanomalies:Geophys J. R. Astr. Sot., v. 12,
p. 151-162.
Bullard, E. C., and Cooper, R. I. B., 1948, The deter-
mination of the masses necessarv to oroduce a eiven
gravitat :ional field: Proc. Roy. Sot. (London) Se;. A.,
194, p. 332-347.
D& pney, C. N. G., 1966a, Three criteria for the judge-
ment of vertical continuation and derivative methods
of geophysical interpretation: Geoexploration, v. 4. p.
3-24.
R= 5
i=l
(gi(Mk)- gY(mJ
Source Technique 51
- 1966b, Geophysical studies in Tasmania; Part
A, Interpretation of the gravitational potential field
using the frequency domain: M.Sc. Thesis (unpubl.),
Dept. of Geology, Univ. of Tasmania.
Danes, 2. F., 1961, Structure calculations from gravity
data and density logs: Mining Trans, v. 223, p. 23-29.
Grant. F. S.. and West. G. F.. 1965. interpretation
theory in applied geophysics: ‘New York, McGraw-
Hill Book Co., Inc.
Hammer, S., 1963, Deep gravity interpretation by
strippmg: Geophysics,v. 28, p. 369-378.
Henderson,R. G., 1960,Comprehensivesystemof auto-
matic computationin magneticand gravity measure-
ments: Geophysics, v. 25, p. X4-585.
Howland-Rose, A., 1966, Derby-Winnaleah Survey,
Tasmania 1964: Dept. of National Development,
Bureau of Mineral Resources, Geology and Geo-
physics, Record No. 1966/10.
Kempthorne, O., 1962, Design and analysis of experi-
ment: New York, John Wiley and Son.
Naudy, H., and Neumann, R., 1965, Sur la dL:finition de
I’anomalie de Bouguer et ses consequencespratiques:
Geophys. Prosp., v. 13, p. l-11.
Neidell! N. S., 1966, Spectral studies of marine geo-
physical profiles: Geophysics, v. 31, p. 122-134.
Nye, P., 1925, The sub-basaltic tin deposits of the
Ringarooma Valley: Tas. Dept. of -Mines, Geological
Survey Bull. No. 35.
Ralston, A., 1965, A first course on numerical analysis:
New York, McGraw-Hill Book Co. Inc.
Roy, A., 1962, Ambiguity in geophysical interpreta-
tion: Geophysics, v. 27, p. 90-99.
Sax, R. L., 1966, Applications of filter theory and infor-
mation theory to the interpretation of gravity mea-
surements: Geophysics, v. 31, p. 570-575.
Strakhov, V. N., and Devitsyn, V. M., 1965, The reduc-
tion of observed values of a potentral field to values at
a constant level: Bull. Acad. Sci. USSR., Geophys.
Ser. (English Transl.), p. 250-255.
Zidarov, D., 1960, Determination des champs de gravi-
tation (ou magnetiques) locaux et regionaux: Comp-
tes rendus de 1’AcadCmie bulgare des Sciences, v. 13,
p. 531-534.
- 1965, Solutions of some inverse problems of ap-
plied geophysics: Geophys. Prosp., v. 13, p. 240-246.
APPENDIX
Derivation of equation (3.12)
Minimize R = (g - AmjT(g - Am) (1)
to find the solution of m in equation (3.12). The
direction of maximum change of R with respect to
rni is given by dR/dm, the ith component of VR
with respect to m.
Thus following Zidarov (1965)
f, (2)
(h - zi)
(3)
 52 Dampney

gi(Mk) is the measured field due to the true masses Developing 4(h) as a power series in X from
Mk, and gij (mk) is the jth approximation of Taylor’s theorem, and taking into consideration
g, from the approximate masses mhj’at (CQ,&, h). only the first two terms (as d3R/ami=0 for all k
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The positions of the rnk are not restricted, but from equation 4), we obtain
for convenience place all the mk at the same
dR dR
height. In the case history, each discrete mass was +(A) = R - X dm ;;
positioned vertically below each of the N data
points making up the survey. However, any other
arrangement is valid within restrictions imposed
(8)
by equation (3.12) and the nature of the matrix
A. In the solution we assume nothing about the d?R
elements of A except that they are real and that __ zz
A is square. dm2
Therefore,
Kow d2j/dm2=0 from equation (a), therefore

f#a(X)= f? - 4Xf2
where
df 4
S
+ 4X2f? _ (10)
k = gk(Mi) - c i=l
( dm ) ’
+(A) will have a maximum when d+/dX=O.
mi(h - zk) Thus
.___-
I (xk - (Yj)’ + (m - pi)’ + (Zk - h)?J3’2 ’
that is
4fz(E)2- Shfl($)” = 0. (11)
S
R = zj;. Hence
(3
z=l

We find following equation (3.11) that

(i) (j-1,
-_x
(j-1)
-__
dR
mk = ‘?& (6)
arn(i-l)
k

where m&j) is a closer approximation to the true

value of mk than rnt-‘).
Therefore,
(12)
aR
R = R m;j’ _ A(‘) -, .’‘
dm(i) ’
which is Zidarov’s (1965) equation (2) with ??I!+
substituted for 8jk and R for LT.
This method assumes R has only one minimum.
The condition for one minimum that d3R/dm3=0
is less than R=R(nz(,“, . , mf$‘).
is seen to be true.
To reduce R as quickly as possible, we maxi-
mize + (A) with respect to X: Geometric considerations

(j) aR Ralston (1965, p. 439-442) derives an equation

+(A) = R&j) - X for X for the case of a symmetric matrix. He also
yjq .‘.)
discusses the geometry of the method of steepest
(I 1 8R descent.
m.v - p_ .
(7) The geometry for the unsymmetric matrix is
am,:: )>
similar. For ideal convergence the magnitude of
 Equivalent Source Technique
Downloaded 01/14/15 to 157.211.3.15. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

FIG. 9. The path of the steepestdescentsolution.

the successive iterations X(l), X(*) . . from the
first approximation is shown in Figure 9.
c=g-Am
in the figure and t represents the major and
minor axes of the hyperellipsoid R.
Physically one sees in the solution of g=Am
that if the true value of WZ~is greatly different
from the other m’s, the hyperellipsoid is very
elongated along the mk axis. This may result if
mk is in a position where the actual source is small
and shallow and the effect described by Bullard
and Cooper (1943) discussed in section 3 occurs.
As the solution will work its way towards
elongated regions of the hyperellipsoid, the in-
fluence of small (relative to the station spacing)
shallow sources will be the last to be extracted
from R. Hence as R is not reduced to zero, the
shallow sources will be treated as noise in the
data. As small shallow sources arc by definition
inadequately sampled, their elimination as noise
is in accordance with normal practice.